test(training): require multi-seed curriculum evaluation

This commit is contained in:
Josh Creek
2026-09-01 17:26:44 +01:00
parent 533ac1afab
commit 9004800326
4 changed files with 32 additions and 4 deletions
+1 -1
View File
@@ -6,7 +6,7 @@ Deferred work, in rough priority order. The current architecture (ShipAction/Shi
The training pipeline is built — see `TRAINING.md` (self-play PPO via the vendored godot_rl_agents bridge, JSON policy export, in-game GDScript inference, eval ladder). Remaining:
- [ ] Run the generation-5 handling/intercepts/league/teamplay curriculum described in `TRAINING.md`; promote later checkpoints as `medium`/`hard` only after they clear the match and behaviour gates.
- [ ] Run the generation-5 handling/intercepts/league/teamplay curriculum described in `TRAINING.md`; promote later checkpoints as `medium`/`hard` only after they clear the match and behaviour gates. The orchestrator now requires three independent paired evaluation seeds for each promotion decision; the current Stage 6 league run remains blocked on its recorded regression/telemetry results.
- [ ] Extend generation 5's moving aerial-intercept states with wall plays and rebound scenarios after Stage 5 establishes a productive-air-touch baseline.
- [ ] Design team-credit rewards and paired 2v2 evaluation before enabling the deferred teamplay stage.